MULTICRITERION OPTIMIZATION OF

LAMINATE STACKING SEQUENCE FOR

MAXIMUM FAILURE MARGINS

Petri Kere 1 and Juhani Koski 2

1

School of Engineering, National University of Uruguay

J. Herrera y Reissig 565, 11300 Montevideo, Uruguay
2
Applied Mechanics, Tampere University of Technology
P.O. Box 589, 33101 Tampere, Finland

SUMMARY: Multicriterion stacking sequence optimization of composite laminates for maximum

failure margins with minimum number of layers is considered. Laminate failure margins to different
loading conditions are measured in terms of laminate initial failure reserve factors that are computed
by applying constant and variable load approach and an iterative line search method. A generalized
failure criterion is used to point out that the internal formulation of the failure criterion has no effect on
the solution procedure. The actual multicriterion optimization problem to be solved is reduced to a
bicriterion problem, where the laminate initial failure reserve factors are combined linearly as the two
strongly competing criteria. Discrete design variables for layer orientation identity are utilized in
creation of laminate symmetric and balanced stacking sequence permutations. The constraint method
is applied to generate maximal solutions. Minimizing distance from the ideal solution in the criterion
space identifies the best-compromise solution.

KEYWORDS: composite laminates, design, multicriterion, multiobjective, optimization, failure

analysis, failure margin

INTRODUCTION
A weight or a volume is a frequently used criterion in the structural optimization. In practical
engineering, however, there are usually several other criteria to measure the performance of a
structure. The criteria may be competing, i.e., an improvement of one criterion can only be achieved
at the cost of another. The naturally competing criteria in designing composite structures are load
carrying capability of a composite laminate against different design loads when the laminate thickness
can only be multiples of the layer thickness. Constant and variable load approach for predicting load
carrying capability of composite laminates gives a quantitative measure on the margin to initial failure

On leave of absence from Helsinki University of Technology. Author to whom correspondence should be
addressed (petri.kere@hut.fi).

[1], [2]. Initial failure is assumed to occur when failure first occurs in one layer or simultaneously in
several layers of the laminate. The laminate initial failure analysis is based on the laminate load
response computed with the Classical Lamination Theory [3]. Margin to initial failure is measured in
terms of a reserve factor RF. In this paper we consider a composite laminate stacking sequence
optimization problem with multiple objectives. Given allowable angles 0, 90, and deg for layer
orientations, find a symmetric and balanced laminate stacking sequence with minimum number of
layers such that laminate initial failure reserve factors due to m loading conditions subjected to the
laminate are maximized. The layer properties as well as the layer thickness t are assumed to be
constant. At least some of the loading conditions are assumed to be conflicting, i.e., the optimum
stacking sequences corresponding to the different loading conditions are different.
The aim of this paper is to study the stacking sequence optimization problem from the multicriterion
optimization point of view. We first represent the laminate analysis formulation appropriate for zeroone programming. In the formulation, zero-one design variables are used to denote the current layer
orientation for each layer of the laminate. The discrete design variables are utilized in creation of
permutations of different stacking sequences of the laminate. Since the formulations of failure criteria
of composite materials may be quite complicated, we apply a derivative-free line search method to
solve the laminate initial failure reserve factors [4]. In this method, failure criterion internal formulation
has no effect on the solution procedure. In the second part of this work, a multicriterion optimization
problem is formulated. The initial problem is reduced to a bicriterion problem, where the vector
objective function consists of laminate reserve factors to two conflicting loading conditions or a linear
combination of reserve factors to several loading conditions. The objective is to obtain a bicriterion
problem with strongly conflicting criteria. The approach has some advantages. The method is efficient
from computational viewpoint yet preserving the multicriterion nature of the original problem. Also
the graphic illustration of maximal solutions in the criterion space is possible. The payoff table by
Benayoun et. al. [5] is utilized in forming the bicriterion problem. We apply the constraint method [6]
for generating maximal solutions in the criterion space. We select the best-compromise solutions by
using the distance function d to minimize the distance from the ideal solution in the criterion space.
A small selection of papers on the laminate stacking sequence optimization by using discrete design
variables comprises [11], [12], [13]. Multicriterion optimization in its present sense originates from
an optimality criterion introduced by Pareto in 1896 [7] for the cases where an improvement of one
criterion can only be achieved at the cost of another. A comprehensive overview on multicriterion
programming can be found in Cohon [6]. Norm methods and partial weighting in multicriterion
optimization of structures is dealt with in Koski and Silvennoinen [8]. Defectiveness of weighting
method in multicriterion optimization of structures is discussed by Koski [9].

COMPOSITE LAMINATE ANALYSIS

Laminate Load Response
A laminate composed of continuous fibre-reinforced-polymer composite layers in different layer
orientations is considered (Fig. 1). Classical Lamination Theory [3] is used to compute properties of
laminates and laminate load response, i.e., the laminate is assumed to be in plane stress state.

Plane of symmetry
1 1
2 2
M
k k
M
N/2 bottom

Fig. 1: Layer numbering convention for laminates.

The constitutive equations of the laminate in plane stress state are
N
A B 0
=
,
M xy B D xy

(1)

Nx
Mx

{N }xy = N y , {M }xy = M y
N
M
xy
xy

(2)

where

are laminate resultant in-plane forces (forces per unit length) and resultant moments (moments per
unit length), respectively, and
0x
x

{0 }xy = 0y , { } xy = y
0

xy
xy

(3)

are the mid-plane strains and the curvatures of the laminate, respectively, in the laminate xyzcoordinate system. Let the laminate be symmetric and balanced, that is, the number of + deg layers
is equal to the deg layers. Due to the symmetry, the laminate coupling stiffness matrix B is
identically zero. By using zero-one design variables
1 if the allowable angle occurs
x =
0 if the allowable angle does not occur

(4)

to represent binary choice for layer orientations, the laminate in-plane and flexural stiffness matrices
A and D may be expressed as
A = 2 Q k t ,
k

D=

2
3

[k

(k 1) 3 Q k t 3 ,

k = 1, 2, K , N 2 ,

in which N is the total number of layers in the laminate and the nonzero elements of Q k are

(5)

+
+
Q11 = U 1 + U 2 ( x k0 x 90
+ c1 x k ) + U 3 ( x k0 + x 90
+ c 2 x k )
k + c1 x k
k + c2 xk

Q 22 = U1 U 2 ( x 0k x k90 + c1 x k+ + c1 x k ) + U 3 ( x k0 + x k90 + c2 x k+ + c 2 x k )
Q 66 = U 5 U 3 ( x k0 + x k90 + c 2 x +k + c 2 x k )
+
Q12 = U 4 U 3 ( x k0 + x 90
+ c 2 x k ) = Q 21
k + c2 xk

(6)

Q16 = 12 U 2 ( c3 x k c 3 x k+ ) U 3 ( c 4 x k c 4 x k+ ) = Q 61
Q 26 = 12 U 2 ( c3 x k c 3 x k+ ) + U 3 (c 4 x k c 4 x k+ ) = Q 62 .
Superscripts 0, 90, +, and denote the corresponding allowable angles for layer orientations and
U1U5 are the stiffness invariants [3]. Constants c1c4 are defined as c 1 =cos2, c 2 =cos4,
c 3 =sin2, and c 4 =sin4. The laminate strain state is achieved with
1

0
A B N
=
.
xy B D M xy

(7)

By computing layer actual strains from the laminate strain state and transforming them to the layer
coordinate system, we get layer equivalent strains. By applying the layer constitutive relations, we get
the layer actual stresses induced by the equivalent strains to be exploited in layer failure prediction.
Laminate Initial Failure Prediction
The laminate is subjected to a set of simultaneously applied loads that may consist of externally
applied mechanical loads, thermal, and moisture induced loads. Each loading condition is partitioned
into a constant and variable load vectors [1], [2]. The basic assumption in the constant and variable
load approach is that the variable load vector, that represents dynamic loads, varies while the
constant load vector representing static pre-loads stays fixed. The probabilistic distributions of the
applied loads and the material strength properties can be taking into account by using appropriate
factors of safety. Factors of safety are defined separately for constant and variable load vectors. The
effective load used in the failure analysis is achieved by multiplying the nominal load with the
corresponding factor of safety.
Let the layer actual stresses in a layer coordinate plane 12 caused by the effective load be .
Superscripts c and v denote the constant and variable load vectors, respectively. In a linear analysis,
the layer stress vector corresponding to the failure load is

= c + RF v ,

(8)

where RFR, RF>0. By using the load criticality factor R, >0, the layer stress vector can be
written as
( ) = c + v .

(9)

The value of a generalized failure criterion F in stress space is obtained from

F ( ) = F [ () ] .

(10)

The failure criterion indicates failure when F ( ) 1. For determining margin to failure, an
unconstrained minimization problem is formulated [4]
min v () = 1 F ( )

[a, b] , R, > 0

(11)

where the objective function is minimized over the closed bounded interval by iteratively reducing the
interval of uncertainty [a, b]. The interval of uncertainty is reduced each time by a factor of the
golden section ratio ( 0.618) until the final length of uncertainty is reached after n iterations. The
point where the failure occurs is determined for each layer as
= ( a n + bn ) 2 RF .

(12)

Laminate failure margin is indicated by the laminate initial failure reserve factor defined as the
minimum of the layer reserve factors
RFL = min RFl t , b , l = 1, 2, K , N ,
l

(13)

where l=1, 2,, N denotes layers of the laminate from top to bottom and superscripts t and b
denote the top and bottom surfaces of the layer. Accordingly, the load criticality factor can be
determined in strain space.

MULTICRITERION PROBLEM
Preliminaries
The problem is formulated as a multicriterion optimization problem
max f ( x) ,

(14)

f ( x ) = { f i ( x) | f i : R, i = 1, 2, K , m} .

(15)

where f is a vector objective function

The components are called criteria and they represent the design objectives by which the
performance of the laminate is measured. The design variables
x = {x k | x k {0,1}, x k = 1, ( x k+ x k ) = 0, k = 1, 2, K , N 2} ,

(16)

where = (0, 90, +, ), belongs to the feasible set defined as

= {x | g ( x) 0, h ( x ) = 0} .

(17)

The image of the feasible set in the criterion space, called the attainable set, is defined by
= {z R m | z = f ( x ), x } .

(18)

Usually, there exists no unique solution which would maximize all m criteria simultaneously. For the
definition of optimal solution of a multicriterion optimization problem we apply the optimality criterion
originally introduced by Pareto [7].
Definition. A solution x* is Pareto optimal for Eqn 14 if and only if there exists no x such that
f i ( x ) f i ( x*) i [1, m]
f i ( x ) > f i ( x*) for at least one i [1, m].

(19)

The points z * = f ( x*) in the criterion spaces are called the maximal points.
Solution Strategy, Criteria, and Constraints
Minimizing the distance from the in general nonfeasible ideal solution in the criterion space by
applying the distance function
d ( z , y ) = max{| z1 y1 |, | z 2 y 2 |}

(20)

optimizes the laminate stacking sequence. First, the optimum initial lay-up configuration is determined
for each loading condition separately. Hence, we get m different lay-up configurations. Laminate
failure margins to the different loading conditions are computed for each laminate and gathered into a
table called the payoff table by Benayoun et. al. [5] (Table 1).
Table 1: Payoff table for m criteria.
f1 (x )
f1 (x 1 )
f1 (x 2 )
M
f1 (x m )

x
x2
M
xm

f2 (x )
f2 (x 1 )
f2 (x 2 )
M
f2 (x m )

L
L
L
M
L

fm (x )
fm (x 1 )
fm (x 2 )
M
fm (x m )

The ideal point is on the diagonal of the payoff table. The minimum value in each column of the
payoff table is the nadir point denoted by n i. To find the most competing criterion, the
normalization coefficients based on the payoff table are computed as
i
f i ( xi ) ni
i =
, where i =
, i = 1, 2, K , m .
max{| f i ( x i ) |, | ni |}
i

(21)

The objective is to obtain a bicriterion problem with strongly competing criteria. The criterion
producing minimum i is selected for the single criterion f 1 (x ) in the new bicriterion problem. The
other criteria are combined linearly as

f 2 ( x ) = i RFi , i > 0, i = 1, i = 1, 2, K , m 1 ,
i

(22)

where weighting coefficients i are selected arbitrarily by the designer. Hence, all the at least rather
non-competing criteria have been combined linearly to yield the problem with the reduced vector
objective function
f ( x ) = { f 1 ( x), f 2 ( x)} .

(23)

The criteria may be non-commensurable, and thus we use a scaled presentation

zi =

f i ( x)
, i = 1, 2 .
max f i

(24)

The feasible set is defined as

= {g i ( x ) = 1 RFi 0, RFi R, RFi > 0, i = 1, 2, K , m} .

(25)

For determining Pareto optimal lay-up configurations permutations of symmetric and balanced
stacking sequences with four allowed layer orientations and fixed initial N are created. Since the
convexity of the attainable set in the criterion space is not guaranteed, we apply the constraint
method [6], [8], [9] to generate maximal solutions. If none of the permutations is feasible, a new set
of laminates is created by adding a layer block {(0, 0), (0, 90), (90, 0), (90, 90), (+, ),
(, +)} deg on the lay-up configuration that minimizes the distance from the ideal solution. Once
again the closest solution to the ideal solution is selected. This procedure is repeated until a lay-up
configuration that satisfies all the constraints is reached. Finally, the number of layers is tried to
reduce, if possible, one layer at a time such that the feasible laminate design is still maintained.
When two conflicting loading conditions are subjected to the laminate, by creating all permutations of
the symmetric and balanced stacking sequences, it is possible to generate all maximal solutions.
When the reduced problem formulation is used, only a subset of maximal solutions is achieved by
fixing the weighting factors. By varying the weighting factors it is possible to generate more maximal
solutions to cover the maximal set more completely.

LAMINATE SUBJECTED TO THREE CONFLICTING LOADING CONDITIONS

The application of a multicriterion stacking sequence optimization problem is illustrated by
considering a laminate that is subjected to three conflicting loading conditions according to the Table
3. The objective is to determine a laminate lay-up configuration with minimum number of layers such
that the laminate initial failure reserve factors are maximized to the loading conditions with FoS = 1.5.
The laminate is composed of layers having the mechanical properties of AS4 Carbon/epoxy ply [10]
(Table 2). The allowable angles for layer orientations are 0, 90, and 45 deg. Tsai-Hill failure
criterion is used.
First, the optimal initial lay-up configuration for each loading condition is determined individually.
Maximization of the laminate initial failure reserve factors to the load cases 1, 2, and 3 with the

minimum allowed number of layers N / 2 = 4 leads to the unique solutions x 1, x 2, and x 3,

respectively. The payoff table is shown in Table 4. Based on the payoff table, 1 = 0.3790,
2 = 0.3139, and 3 = 0.3071.
Table 2: Mechanical properties of the ply.
Ply AS4 Carbon / epoxy (AS4 / 3501-6) t = 0.25 mm
Ply engineering constants
Ply failure stresses
E1 = 126 GPa
Xt = 1950 MPa
E2 = 11 GPa
Yt = 48 MPa
G12 = 6.6 GPa
S 12 = 79 MPa
12 = 0.28

Xc = 1480 MPa
Yc = 200 MPa

Table 3: The three conflicting loading conditions.

Load case 1
Nx = 55 kN/m
Ny = 0
Nxy = 100 kN/m
Mx = 10 Nm/m
My = 0
Mxy = 10 Nm/m

Load case 2
Nx = 50 kN/m
Ny = 260 kN/m
Nxy = 0
Mx = 0
My = 10 Nm/m
Mxy = 0

Load case 3
Nx = 280 kN/m
Ny = 50 kN/m
Nxy = 0
Mx = 10 Nm/m
My = 0
Mxy = 0

Table 4: The Payoff table for the laminate design problem.

x
x 1 = [+45/ 45/0/90]SE
x 2 = [90/90/90/0]SE
x 3 = [0/0/0/90]SE

RF1 (x)
2.8326
0.8030
0.8101

RF2 (x)
2.0826
5.1215
2.1853

RF3 (x)
2.0172
2.0415
4.8081

Therefore, the problem with the reduced vector objective function is formulated as
max {RF3 ( x ), 1 RF1 ( x) + (1 1 ) RF 2 ( x )}
x

(26)

= {g i ( x) = 1 RFi 0, RFi R, RFi > 0, i = 1, 2, 3}.

Table 5: Pareto optimal lay-up configurations at the maximal points to the case 1 = 0.5
shown in Fig. 1. The best-compromise solution is underlined.
Lay-up
[90/+45/ 45/90]SE
[90/90/+45/ 45]SE
[+45/90/ 45/0]SE
[+45/90/0/ 45]SE
[90/+45/0/ 45]SE
[+45/0/90/ 45]SE
[90/0/+45/ 45/]SE
[0/90/+45/ 45]SE
[+45/ 45/0/0]SE
[+45/0/ 45/0]SE
[+45/0/0/ 45]SE
[0/+45/ 45/0]SE
[0/+45/0/ 45]SE

z1
0.2681
0.2694
0.5208
0.5471
0.5679
0.5824
0.6139
0.6270
0.7872
0.8703
0.9008
0.9529
0.9659

z2
1
0.9776
0.7940
0.7789
0.7556
0.7370
0.7149
0.7020
0.5722
0.5718
0.5368
0.5336
0.4967

RF1
2.4228
2.0690
2.7212
2.5952
2.3007
2.4828
1.9922
1.9678
2.5868
2.5834
2.3845
2.3135
2.1011

RF2
3.9431
4.1545
2.3332
2.3631
2.5092
2.2091
2.5589
2.5009
1.0557
1.0564
1.0327
1.0831
1.0611

RF3
1.0080
1.0132
1.9585
2.0573
2.1354
2.1901
2.3084
2.3576
2.9603
3.2727
3.3874
3.5834
3.6320

d
0.7319
0.7306
0.4792
0.4529
0.4321
0.4176
0.3861
0.3730
0.4278
0.4282
0.4632
0.4664
0.5033

[0/0/+45/ 45]SE

0.4756

1.9385

1.0890

3.7604

0.5244

Permutations of symmetric and balanced stacking sequences with allowable layer orientations are
created. The maximal solutions in bicriterion problem are computed by using the constraint method.
The maximal points achieved with 1 =1, 0.75, 0.5, 0.25, and 0 are shown in Fig. 2. Pareto optimal
laminate lay-up configurations corresponding to 1 = 0.5 are given numerically in Table 5. This is the
case, in which the weighting coefficients of failure margins to the both load cases are equal.

1
1

0.9
0.75

0.8

z2

0.7
0.5

0.6
0.5

0.25

0.4
0.3
0.2
0.2

0.3

0.4

0.5

0.6
z

0.7

0.8

0.9

Fig. 2:

Maximal points to the bicriterion problem with three conflicting loading conditions,
1 = 1, 0.75, 0.5, 0.25, and 0. The best-compromise solution to the case 1 = 0.5 is
marked with the star.

CONCLUSIONS
Generating Pareto optimal solutions for a multicriterion stacking sequence optimization problem of
composite laminates subjected to multiple conflicting loading conditions is presented. Laminate initial
failure reserve factors are treated as criteria and computed with an iterative line search method. The
method is suitable to be used also with complicated failure criterion formulations, since the internal
formulation of the failure criterion has no effect on the solution procedure. The optimization scheme is
based on the solution of bicriterion subproblems, which enables the designer to obtain results in any
accuracy that is relevant. Using the reduced vector objective function is computationally efficient still
preserving the multicriterion nature of the problem. Also the graphical illustration of the maximal
solutions in the criterion space is possible.

ACKNOWLEDGEMENT
The work of the first author was supported by the Academy of Finland Grant 65375 and the
Sanoma Oy foundation (Helsingin Sanomain 100-vuotissti).

REFERENCES
1.

2.

3.
4.

5.

6.
7.
8.
9.
10.
11.
12.

13.

Palanter, M. and Klein, M., Constant and Variable Loads in Failure Analyses of
Composite Laminates, Computer Aided Design in Composite Material Technology IV,
Southampton, UK, 1994, Blain, W. R. and De Wilde, W. P., Eds, Computational
Mechanics Publications, pp. 221228.
Palanter, M. and Karjalainen, J. P., Failure Margins of Composite Laminates with
Constant and Variable Load Approach, ECCM-8, European Conference on Composite
Materials, Naples, Italy, June 3-6, 1998, Vol. 3, Visconti, C., Ed., Woodhead Publishing
Limited, pp. 5158.
Jones, R. M., Mechanics of Composite Materials, Hemisphere Publishing Corporation,
New York, 1975.
Kere, P. and Palanter, M., A Method for Solving Margins of Safety in Composite Failure
Analysis, Proceedings of the 6th Finnish Mechanics Days, Oulu, Finland, 1997,
University of Oulu, Structural Engineering Laboratory, Publication 56, pp.187197.
Benayoun, R., de Montgolfier, J., Tergny, J., and Larichev, O., Linear Programming with
Multiple Objective Functions: STEP Method (STEM), Mathl. Program., Vol. 1, 1971,
pp. 366375.
Cohon, J. L., Multiobjective Programming and Planning, Mathematics in Science and
Engineering, Vol. 140, Academic Press, Inc., New York, 1978.
Pareto, V., Manual of Political Economy, The Macmillan Press, New York, 1971.
Koski, J. and Silvennoinen, R., Norm Methods and Partial Weighting in Multicriterion
Optimization of Structures, Int. J. Numer. Methods Eng., Vol. 24, 1987, pp. 11011121.
Koski, J., Defectiviness of Weighting Method in Multicriterion Optimization of Structures,
Communications in Applied Numerical Methods, Vol. 1, 1985, pp. 333337.
Hinton, M. J., Soden, P. D., and Kaddour, A. S., Failure Criteria in Fibre-ReinforcedPolymer Composites Composites Science and Technology, Vol. 58, 1998.
Haftka, R. T. and Walsh, J. L., Stacking Sequence Optimization for Buckling of Laminated
Plates by Integer Programming, AIAA Journal, Vol. 30, 1992, pp. 814819.
Nagendra, S., Haftka, R. T., and Grdal, Z., Stacking Sequence Optimization of Simply
Supported Laminates with Stability and Strain Constraints, AIAA Journal, Vol. 30, 1992,
pp. 21322137.
Le Riche, R. and Haftka, R. T., Optimization of Laminate Stacking Sequence for Buckling
Load Maximization by Genetic Algorithm, AIAA Journal, Vol. 31, 1993, pp. 951956.